2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
53 #include <asm/processor.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
58 #include "coalesced_mmio.h"
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
70 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
73 DEFINE_RAW_SPINLOCK(kvm_lock
);
76 static cpumask_var_t cpus_hardware_enabled
;
77 static int kvm_usage_count
= 0;
78 static atomic_t hardware_enable_failed
;
80 struct kmem_cache
*kvm_vcpu_cache
;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
83 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
85 struct dentry
*kvm_debugfs_dir
;
87 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
90 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
96 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
99 EXPORT_SYMBOL_GPL(kvm_rebooting
);
101 static bool largepages_enabled
= true;
103 bool kvm_is_mmio_pfn(pfn_t pfn
)
105 if (pfn_valid(pfn
)) {
107 struct page
*tail
= pfn_to_page(pfn
);
108 struct page
*head
= compound_trans_head(tail
);
109 reserved
= PageReserved(head
);
112 * "head" is not a dangling pointer
113 * (compound_trans_head takes care of that)
114 * but the hugepage may have been splitted
115 * from under us (and we may not hold a
116 * reference count on the head page so it can
117 * be reused before we run PageReferenced), so
118 * we've to check PageTail before returning
125 return PageReserved(tail
);
132 * Switches to specified vcpu, until a matching vcpu_put()
134 int vcpu_load(struct kvm_vcpu
*vcpu
)
138 if (mutex_lock_killable(&vcpu
->mutex
))
140 if (unlikely(vcpu
->pid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
141 /* The thread running this VCPU changed. */
142 struct pid
*oldpid
= vcpu
->pid
;
143 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
144 rcu_assign_pointer(vcpu
->pid
, newpid
);
149 preempt_notifier_register(&vcpu
->preempt_notifier
);
150 kvm_arch_vcpu_load(vcpu
, cpu
);
155 void vcpu_put(struct kvm_vcpu
*vcpu
)
158 kvm_arch_vcpu_put(vcpu
);
159 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
161 mutex_unlock(&vcpu
->mutex
);
164 static void ack_flush(void *_completed
)
168 static bool make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
173 struct kvm_vcpu
*vcpu
;
175 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
178 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
179 kvm_make_request(req
, vcpu
);
182 /* Set ->requests bit before we read ->mode */
185 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
186 kvm_vcpu_exiting_guest_mode(vcpu
) != OUTSIDE_GUEST_MODE
)
187 cpumask_set_cpu(cpu
, cpus
);
189 if (unlikely(cpus
== NULL
))
190 smp_call_function_many(cpu_online_mask
, ack_flush
, NULL
, 1);
191 else if (!cpumask_empty(cpus
))
192 smp_call_function_many(cpus
, ack_flush
, NULL
, 1);
196 free_cpumask_var(cpus
);
200 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
202 long dirty_count
= kvm
->tlbs_dirty
;
205 if (make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
206 ++kvm
->stat
.remote_tlb_flush
;
207 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
210 void kvm_reload_remote_mmus(struct kvm
*kvm
)
212 make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
215 void kvm_make_mclock_inprogress_request(struct kvm
*kvm
)
217 make_all_cpus_request(kvm
, KVM_REQ_MCLOCK_INPROGRESS
);
220 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
225 mutex_init(&vcpu
->mutex
);
230 init_waitqueue_head(&vcpu
->wq
);
231 kvm_async_pf_vcpu_init(vcpu
);
233 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
238 vcpu
->run
= page_address(page
);
240 kvm_vcpu_set_in_spin_loop(vcpu
, false);
241 kvm_vcpu_set_dy_eligible(vcpu
, false);
243 r
= kvm_arch_vcpu_init(vcpu
);
249 free_page((unsigned long)vcpu
->run
);
253 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
255 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
258 kvm_arch_vcpu_uninit(vcpu
);
259 free_page((unsigned long)vcpu
->run
);
261 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
263 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
264 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
266 return container_of(mn
, struct kvm
, mmu_notifier
);
269 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier
*mn
,
270 struct mm_struct
*mm
,
271 unsigned long address
)
273 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
274 int need_tlb_flush
, idx
;
277 * When ->invalidate_page runs, the linux pte has been zapped
278 * already but the page is still allocated until
279 * ->invalidate_page returns. So if we increase the sequence
280 * here the kvm page fault will notice if the spte can't be
281 * established because the page is going to be freed. If
282 * instead the kvm page fault establishes the spte before
283 * ->invalidate_page runs, kvm_unmap_hva will release it
286 * The sequence increase only need to be seen at spin_unlock
287 * time, and not at spin_lock time.
289 * Increasing the sequence after the spin_unlock would be
290 * unsafe because the kvm page fault could then establish the
291 * pte after kvm_unmap_hva returned, without noticing the page
292 * is going to be freed.
294 idx
= srcu_read_lock(&kvm
->srcu
);
295 spin_lock(&kvm
->mmu_lock
);
297 kvm
->mmu_notifier_seq
++;
298 need_tlb_flush
= kvm_unmap_hva(kvm
, address
) | kvm
->tlbs_dirty
;
299 /* we've to flush the tlb before the pages can be freed */
301 kvm_flush_remote_tlbs(kvm
);
303 spin_unlock(&kvm
->mmu_lock
);
304 srcu_read_unlock(&kvm
->srcu
, idx
);
307 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
308 struct mm_struct
*mm
,
309 unsigned long address
,
312 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
315 idx
= srcu_read_lock(&kvm
->srcu
);
316 spin_lock(&kvm
->mmu_lock
);
317 kvm
->mmu_notifier_seq
++;
318 kvm_set_spte_hva(kvm
, address
, pte
);
319 spin_unlock(&kvm
->mmu_lock
);
320 srcu_read_unlock(&kvm
->srcu
, idx
);
323 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
324 struct mm_struct
*mm
,
328 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
329 int need_tlb_flush
= 0, idx
;
331 idx
= srcu_read_lock(&kvm
->srcu
);
332 spin_lock(&kvm
->mmu_lock
);
334 * The count increase must become visible at unlock time as no
335 * spte can be established without taking the mmu_lock and
336 * count is also read inside the mmu_lock critical section.
338 kvm
->mmu_notifier_count
++;
339 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
340 need_tlb_flush
|= kvm
->tlbs_dirty
;
341 /* we've to flush the tlb before the pages can be freed */
343 kvm_flush_remote_tlbs(kvm
);
345 spin_unlock(&kvm
->mmu_lock
);
346 srcu_read_unlock(&kvm
->srcu
, idx
);
349 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
350 struct mm_struct
*mm
,
354 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
356 spin_lock(&kvm
->mmu_lock
);
358 * This sequence increase will notify the kvm page fault that
359 * the page that is going to be mapped in the spte could have
362 kvm
->mmu_notifier_seq
++;
365 * The above sequence increase must be visible before the
366 * below count decrease, which is ensured by the smp_wmb above
367 * in conjunction with the smp_rmb in mmu_notifier_retry().
369 kvm
->mmu_notifier_count
--;
370 spin_unlock(&kvm
->mmu_lock
);
372 BUG_ON(kvm
->mmu_notifier_count
< 0);
375 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
376 struct mm_struct
*mm
,
377 unsigned long address
)
379 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
382 idx
= srcu_read_lock(&kvm
->srcu
);
383 spin_lock(&kvm
->mmu_lock
);
385 young
= kvm_age_hva(kvm
, address
);
387 kvm_flush_remote_tlbs(kvm
);
389 spin_unlock(&kvm
->mmu_lock
);
390 srcu_read_unlock(&kvm
->srcu
, idx
);
395 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
396 struct mm_struct
*mm
,
397 unsigned long address
)
399 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
402 idx
= srcu_read_lock(&kvm
->srcu
);
403 spin_lock(&kvm
->mmu_lock
);
404 young
= kvm_test_age_hva(kvm
, address
);
405 spin_unlock(&kvm
->mmu_lock
);
406 srcu_read_unlock(&kvm
->srcu
, idx
);
411 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
412 struct mm_struct
*mm
)
414 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
417 idx
= srcu_read_lock(&kvm
->srcu
);
418 kvm_arch_flush_shadow_all(kvm
);
419 srcu_read_unlock(&kvm
->srcu
, idx
);
422 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
423 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
424 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
425 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
426 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
427 .test_young
= kvm_mmu_notifier_test_young
,
428 .change_pte
= kvm_mmu_notifier_change_pte
,
429 .release
= kvm_mmu_notifier_release
,
432 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
434 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
435 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
438 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
440 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
445 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
447 static void kvm_init_memslots_id(struct kvm
*kvm
)
450 struct kvm_memslots
*slots
= kvm
->memslots
;
452 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
453 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
456 static struct kvm
*kvm_create_vm(unsigned long type
)
459 struct kvm
*kvm
= kvm_arch_alloc_vm();
462 return ERR_PTR(-ENOMEM
);
464 r
= kvm_arch_init_vm(kvm
, type
);
466 goto out_err_nodisable
;
468 r
= hardware_enable_all();
470 goto out_err_nodisable
;
472 #ifdef CONFIG_HAVE_KVM_IRQCHIP
473 INIT_HLIST_HEAD(&kvm
->mask_notifier_list
);
474 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
478 kvm
->memslots
= kzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
481 kvm_init_memslots_id(kvm
);
482 if (init_srcu_struct(&kvm
->srcu
))
484 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
485 kvm
->buses
[i
] = kzalloc(sizeof(struct kvm_io_bus
),
491 spin_lock_init(&kvm
->mmu_lock
);
492 kvm
->mm
= current
->mm
;
493 atomic_inc(&kvm
->mm
->mm_count
);
494 kvm_eventfd_init(kvm
);
495 mutex_init(&kvm
->lock
);
496 mutex_init(&kvm
->irq_lock
);
497 mutex_init(&kvm
->slots_lock
);
498 atomic_set(&kvm
->users_count
, 1);
500 r
= kvm_init_mmu_notifier(kvm
);
504 raw_spin_lock(&kvm_lock
);
505 list_add(&kvm
->vm_list
, &vm_list
);
506 raw_spin_unlock(&kvm_lock
);
511 cleanup_srcu_struct(&kvm
->srcu
);
513 hardware_disable_all();
515 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
516 kfree(kvm
->buses
[i
]);
517 kfree(kvm
->memslots
);
518 kvm_arch_free_vm(kvm
);
523 * Avoid using vmalloc for a small buffer.
524 * Should not be used when the size is statically known.
526 void *kvm_kvzalloc(unsigned long size
)
528 if (size
> PAGE_SIZE
)
529 return vzalloc(size
);
531 return kzalloc(size
, GFP_KERNEL
);
534 void kvm_kvfree(const void *addr
)
536 if (is_vmalloc_addr(addr
))
542 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
544 if (!memslot
->dirty_bitmap
)
547 kvm_kvfree(memslot
->dirty_bitmap
);
548 memslot
->dirty_bitmap
= NULL
;
552 * Free any memory in @free but not in @dont.
554 static void kvm_free_physmem_slot(struct kvm_memory_slot
*free
,
555 struct kvm_memory_slot
*dont
)
557 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
558 kvm_destroy_dirty_bitmap(free
);
560 kvm_arch_free_memslot(free
, dont
);
565 void kvm_free_physmem(struct kvm
*kvm
)
567 struct kvm_memslots
*slots
= kvm
->memslots
;
568 struct kvm_memory_slot
*memslot
;
570 kvm_for_each_memslot(memslot
, slots
)
571 kvm_free_physmem_slot(memslot
, NULL
);
573 kfree(kvm
->memslots
);
576 static void kvm_destroy_vm(struct kvm
*kvm
)
579 struct mm_struct
*mm
= kvm
->mm
;
581 kvm_arch_sync_events(kvm
);
582 raw_spin_lock(&kvm_lock
);
583 list_del(&kvm
->vm_list
);
584 raw_spin_unlock(&kvm_lock
);
585 kvm_free_irq_routing(kvm
);
586 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
587 kvm_io_bus_destroy(kvm
->buses
[i
]);
588 kvm_coalesced_mmio_free(kvm
);
589 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
590 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
592 kvm_arch_flush_shadow_all(kvm
);
594 kvm_arch_destroy_vm(kvm
);
595 kvm_free_physmem(kvm
);
596 cleanup_srcu_struct(&kvm
->srcu
);
597 kvm_arch_free_vm(kvm
);
598 hardware_disable_all();
602 void kvm_get_kvm(struct kvm
*kvm
)
604 atomic_inc(&kvm
->users_count
);
606 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
608 void kvm_put_kvm(struct kvm
*kvm
)
610 if (atomic_dec_and_test(&kvm
->users_count
))
613 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
616 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
618 struct kvm
*kvm
= filp
->private_data
;
620 kvm_irqfd_release(kvm
);
627 * Allocation size is twice as large as the actual dirty bitmap size.
628 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
630 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
633 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
635 memslot
->dirty_bitmap
= kvm_kvzalloc(dirty_bytes
);
636 if (!memslot
->dirty_bitmap
)
639 #endif /* !CONFIG_S390 */
643 static int cmp_memslot(const void *slot1
, const void *slot2
)
645 struct kvm_memory_slot
*s1
, *s2
;
647 s1
= (struct kvm_memory_slot
*)slot1
;
648 s2
= (struct kvm_memory_slot
*)slot2
;
650 if (s1
->npages
< s2
->npages
)
652 if (s1
->npages
> s2
->npages
)
659 * Sort the memslots base on its size, so the larger slots
660 * will get better fit.
662 static void sort_memslots(struct kvm_memslots
*slots
)
666 sort(slots
->memslots
, KVM_MEM_SLOTS_NUM
,
667 sizeof(struct kvm_memory_slot
), cmp_memslot
, NULL
);
669 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
670 slots
->id_to_index
[slots
->memslots
[i
].id
] = i
;
673 void update_memslots(struct kvm_memslots
*slots
, struct kvm_memory_slot
*new)
677 struct kvm_memory_slot
*old
= id_to_memslot(slots
, id
);
678 unsigned long npages
= old
->npages
;
681 if (new->npages
!= npages
)
682 sort_memslots(slots
);
688 static int check_memory_region_flags(struct kvm_userspace_memory_region
*mem
)
690 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
692 #ifdef KVM_CAP_READONLY_MEM
693 valid_flags
|= KVM_MEM_READONLY
;
696 if (mem
->flags
& ~valid_flags
)
703 * Allocate some memory and give it an address in the guest physical address
706 * Discontiguous memory is allowed, mostly for framebuffers.
708 * Must be called holding mmap_sem for write.
710 int __kvm_set_memory_region(struct kvm
*kvm
,
711 struct kvm_userspace_memory_region
*mem
,
716 unsigned long npages
;
718 struct kvm_memory_slot
*memslot
;
719 struct kvm_memory_slot old
, new;
720 struct kvm_memslots
*slots
, *old_memslots
;
722 r
= check_memory_region_flags(mem
);
727 /* General sanity checks */
728 if (mem
->memory_size
& (PAGE_SIZE
- 1))
730 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
732 /* We can read the guest memory with __xxx_user() later on. */
734 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
735 !access_ok(VERIFY_WRITE
,
736 (void __user
*)(unsigned long)mem
->userspace_addr
,
739 if (mem
->slot
>= KVM_MEM_SLOTS_NUM
)
741 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
744 memslot
= id_to_memslot(kvm
->memslots
, mem
->slot
);
745 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
746 npages
= mem
->memory_size
>> PAGE_SHIFT
;
749 if (npages
> KVM_MEM_MAX_NR_PAGES
)
753 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
755 new = old
= *memslot
;
758 new.base_gfn
= base_gfn
;
760 new.flags
= mem
->flags
;
762 /* Disallow changing a memory slot's size. */
764 if (npages
&& old
.npages
&& npages
!= old
.npages
)
767 /* Check for overlaps */
769 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
770 struct kvm_memory_slot
*s
= &kvm
->memslots
->memslots
[i
];
772 if (s
== memslot
|| !s
->npages
)
774 if (!((base_gfn
+ npages
<= s
->base_gfn
) ||
775 (base_gfn
>= s
->base_gfn
+ s
->npages
)))
779 /* Free page dirty bitmap if unneeded */
780 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
781 new.dirty_bitmap
= NULL
;
785 /* Allocate if a slot is being created */
786 if (npages
&& !old
.npages
) {
787 new.user_alloc
= user_alloc
;
788 new.userspace_addr
= mem
->userspace_addr
;
790 if (kvm_arch_create_memslot(&new, npages
))
794 /* Allocate page dirty bitmap if needed */
795 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
796 if (kvm_create_dirty_bitmap(&new) < 0)
798 /* destroy any largepage mappings for dirty tracking */
801 if (!npages
|| base_gfn
!= old
.base_gfn
) {
802 struct kvm_memory_slot
*slot
;
805 slots
= kmemdup(kvm
->memslots
, sizeof(struct kvm_memslots
),
809 slot
= id_to_memslot(slots
, mem
->slot
);
810 slot
->flags
|= KVM_MEMSLOT_INVALID
;
812 update_memslots(slots
, NULL
);
814 old_memslots
= kvm
->memslots
;
815 rcu_assign_pointer(kvm
->memslots
, slots
);
816 synchronize_srcu_expedited(&kvm
->srcu
);
817 /* From this point no new shadow pages pointing to a deleted,
818 * or moved, memslot will be created.
820 * validation of sp->gfn happens in:
821 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
822 * - kvm_is_visible_gfn (mmu_check_roots)
824 kvm_arch_flush_shadow_memslot(kvm
, slot
);
828 r
= kvm_arch_prepare_memory_region(kvm
, &new, old
, mem
, user_alloc
);
832 /* map/unmap the pages in iommu page table */
834 r
= kvm_iommu_map_pages(kvm
, &new);
838 kvm_iommu_unmap_pages(kvm
, &old
);
841 slots
= kmemdup(kvm
->memslots
, sizeof(struct kvm_memslots
),
846 /* actual memory is freed via old in kvm_free_physmem_slot below */
848 new.dirty_bitmap
= NULL
;
849 memset(&new.arch
, 0, sizeof(new.arch
));
852 update_memslots(slots
, &new);
853 old_memslots
= kvm
->memslots
;
854 rcu_assign_pointer(kvm
->memslots
, slots
);
855 synchronize_srcu_expedited(&kvm
->srcu
);
857 kvm_arch_commit_memory_region(kvm
, mem
, old
, user_alloc
);
859 kvm_free_physmem_slot(&old
, &new);
865 kvm_free_physmem_slot(&new, &old
);
870 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
872 int kvm_set_memory_region(struct kvm
*kvm
,
873 struct kvm_userspace_memory_region
*mem
,
878 mutex_lock(&kvm
->slots_lock
);
879 r
= __kvm_set_memory_region(kvm
, mem
, user_alloc
);
880 mutex_unlock(&kvm
->slots_lock
);
883 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
885 int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
887 kvm_userspace_memory_region
*mem
,
890 if (mem
->slot
>= KVM_MEMORY_SLOTS
)
892 return kvm_set_memory_region(kvm
, mem
, user_alloc
);
895 int kvm_get_dirty_log(struct kvm
*kvm
,
896 struct kvm_dirty_log
*log
, int *is_dirty
)
898 struct kvm_memory_slot
*memslot
;
901 unsigned long any
= 0;
904 if (log
->slot
>= KVM_MEMORY_SLOTS
)
907 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
909 if (!memslot
->dirty_bitmap
)
912 n
= kvm_dirty_bitmap_bytes(memslot
);
914 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
915 any
= memslot
->dirty_bitmap
[i
];
918 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
929 bool kvm_largepages_enabled(void)
931 return largepages_enabled
;
934 void kvm_disable_largepages(void)
936 largepages_enabled
= false;
938 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
940 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
942 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
944 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
946 int kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
948 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
950 if (!memslot
|| memslot
->id
>= KVM_MEMORY_SLOTS
||
951 memslot
->flags
& KVM_MEMSLOT_INVALID
)
956 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
958 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
960 struct vm_area_struct
*vma
;
961 unsigned long addr
, size
;
965 addr
= gfn_to_hva(kvm
, gfn
);
966 if (kvm_is_error_hva(addr
))
969 down_read(¤t
->mm
->mmap_sem
);
970 vma
= find_vma(current
->mm
, addr
);
974 size
= vma_kernel_pagesize(vma
);
977 up_read(¤t
->mm
->mmap_sem
);
982 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
984 return slot
->flags
& KVM_MEM_READONLY
;
987 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
988 gfn_t
*nr_pages
, bool write
)
990 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
991 return KVM_HVA_ERR_BAD
;
993 if (memslot_is_readonly(slot
) && write
)
994 return KVM_HVA_ERR_RO_BAD
;
997 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
999 return __gfn_to_hva_memslot(slot
, gfn
);
1002 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1005 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1008 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1011 return gfn_to_hva_many(slot
, gfn
, NULL
);
1013 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1015 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1017 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1019 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1022 * The hva returned by this function is only allowed to be read.
1023 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1025 static unsigned long gfn_to_hva_read(struct kvm
*kvm
, gfn_t gfn
)
1027 return __gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
, false);
1030 static int kvm_read_hva(void *data
, void __user
*hva
, int len
)
1032 return __copy_from_user(data
, hva
, len
);
1035 static int kvm_read_hva_atomic(void *data
, void __user
*hva
, int len
)
1037 return __copy_from_user_inatomic(data
, hva
, len
);
1040 int get_user_page_nowait(struct task_struct
*tsk
, struct mm_struct
*mm
,
1041 unsigned long start
, int write
, struct page
**page
)
1043 int flags
= FOLL_TOUCH
| FOLL_NOWAIT
| FOLL_HWPOISON
| FOLL_GET
;
1046 flags
|= FOLL_WRITE
;
1048 return __get_user_pages(tsk
, mm
, start
, 1, flags
, page
, NULL
, NULL
);
1051 static inline int check_user_page_hwpoison(unsigned long addr
)
1053 int rc
, flags
= FOLL_TOUCH
| FOLL_HWPOISON
| FOLL_WRITE
;
1055 rc
= __get_user_pages(current
, current
->mm
, addr
, 1,
1056 flags
, NULL
, NULL
, NULL
);
1057 return rc
== -EHWPOISON
;
1061 * The atomic path to get the writable pfn which will be stored in @pfn,
1062 * true indicates success, otherwise false is returned.
1064 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1065 bool write_fault
, bool *writable
, pfn_t
*pfn
)
1067 struct page
*page
[1];
1070 if (!(async
|| atomic
))
1074 * Fast pin a writable pfn only if it is a write fault request
1075 * or the caller allows to map a writable pfn for a read fault
1078 if (!(write_fault
|| writable
))
1081 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1083 *pfn
= page_to_pfn(page
[0]);
1094 * The slow path to get the pfn of the specified host virtual address,
1095 * 1 indicates success, -errno is returned if error is detected.
1097 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1098 bool *writable
, pfn_t
*pfn
)
1100 struct page
*page
[1];
1106 *writable
= write_fault
;
1109 down_read(¤t
->mm
->mmap_sem
);
1110 npages
= get_user_page_nowait(current
, current
->mm
,
1111 addr
, write_fault
, page
);
1112 up_read(¤t
->mm
->mmap_sem
);
1114 npages
= get_user_pages_fast(addr
, 1, write_fault
,
1119 /* map read fault as writable if possible */
1120 if (unlikely(!write_fault
) && writable
) {
1121 struct page
*wpage
[1];
1123 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1132 *pfn
= page_to_pfn(page
[0]);
1136 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1138 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1141 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1148 * Pin guest page in memory and return its pfn.
1149 * @addr: host virtual address which maps memory to the guest
1150 * @atomic: whether this function can sleep
1151 * @async: whether this function need to wait IO complete if the
1152 * host page is not in the memory
1153 * @write_fault: whether we should get a writable host page
1154 * @writable: whether it allows to map a writable host page for !@write_fault
1156 * The function will map a writable host page for these two cases:
1157 * 1): @write_fault = true
1158 * 2): @write_fault = false && @writable, @writable will tell the caller
1159 * whether the mapping is writable.
1161 static pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1162 bool write_fault
, bool *writable
)
1164 struct vm_area_struct
*vma
;
1168 /* we can do it either atomically or asynchronously, not both */
1169 BUG_ON(atomic
&& async
);
1171 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1175 return KVM_PFN_ERR_FAULT
;
1177 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1181 down_read(¤t
->mm
->mmap_sem
);
1182 if (npages
== -EHWPOISON
||
1183 (!async
&& check_user_page_hwpoison(addr
))) {
1184 pfn
= KVM_PFN_ERR_HWPOISON
;
1188 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1191 pfn
= KVM_PFN_ERR_FAULT
;
1192 else if ((vma
->vm_flags
& VM_PFNMAP
)) {
1193 pfn
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) +
1195 BUG_ON(!kvm_is_mmio_pfn(pfn
));
1197 if (async
&& vma_is_valid(vma
, write_fault
))
1199 pfn
= KVM_PFN_ERR_FAULT
;
1202 up_read(¤t
->mm
->mmap_sem
);
1207 __gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
, bool atomic
,
1208 bool *async
, bool write_fault
, bool *writable
)
1210 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1212 if (addr
== KVM_HVA_ERR_RO_BAD
)
1213 return KVM_PFN_ERR_RO_FAULT
;
1215 if (kvm_is_error_hva(addr
))
1216 return KVM_PFN_NOSLOT
;
1218 /* Do not map writable pfn in the readonly memslot. */
1219 if (writable
&& memslot_is_readonly(slot
)) {
1224 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1228 static pfn_t
__gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
, bool atomic
, bool *async
,
1229 bool write_fault
, bool *writable
)
1231 struct kvm_memory_slot
*slot
;
1236 slot
= gfn_to_memslot(kvm
, gfn
);
1238 return __gfn_to_pfn_memslot(slot
, gfn
, atomic
, async
, write_fault
,
1242 pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1244 return __gfn_to_pfn(kvm
, gfn
, true, NULL
, true, NULL
);
1246 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1248 pfn_t
gfn_to_pfn_async(struct kvm
*kvm
, gfn_t gfn
, bool *async
,
1249 bool write_fault
, bool *writable
)
1251 return __gfn_to_pfn(kvm
, gfn
, false, async
, write_fault
, writable
);
1253 EXPORT_SYMBOL_GPL(gfn_to_pfn_async
);
1255 pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1257 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, true, NULL
);
1259 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1261 pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1264 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, write_fault
, writable
);
1266 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1268 pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1270 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1273 pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1275 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1277 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1279 int gfn_to_page_many_atomic(struct kvm
*kvm
, gfn_t gfn
, struct page
**pages
,
1285 addr
= gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, &entry
);
1286 if (kvm_is_error_hva(addr
))
1289 if (entry
< nr_pages
)
1292 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1294 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1296 static struct page
*kvm_pfn_to_page(pfn_t pfn
)
1298 if (is_error_noslot_pfn(pfn
))
1299 return KVM_ERR_PTR_BAD_PAGE
;
1301 if (kvm_is_mmio_pfn(pfn
)) {
1303 return KVM_ERR_PTR_BAD_PAGE
;
1306 return pfn_to_page(pfn
);
1309 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1313 pfn
= gfn_to_pfn(kvm
, gfn
);
1315 return kvm_pfn_to_page(pfn
);
1318 EXPORT_SYMBOL_GPL(gfn_to_page
);
1320 void kvm_release_page_clean(struct page
*page
)
1322 WARN_ON(is_error_page(page
));
1324 kvm_release_pfn_clean(page_to_pfn(page
));
1326 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1328 void kvm_release_pfn_clean(pfn_t pfn
)
1330 if (!is_error_noslot_pfn(pfn
) && !kvm_is_mmio_pfn(pfn
))
1331 put_page(pfn_to_page(pfn
));
1333 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1335 void kvm_release_page_dirty(struct page
*page
)
1337 WARN_ON(is_error_page(page
));
1339 kvm_release_pfn_dirty(page_to_pfn(page
));
1341 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1343 void kvm_release_pfn_dirty(pfn_t pfn
)
1345 kvm_set_pfn_dirty(pfn
);
1346 kvm_release_pfn_clean(pfn
);
1348 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
1350 void kvm_set_page_dirty(struct page
*page
)
1352 kvm_set_pfn_dirty(page_to_pfn(page
));
1354 EXPORT_SYMBOL_GPL(kvm_set_page_dirty
);
1356 void kvm_set_pfn_dirty(pfn_t pfn
)
1358 if (!kvm_is_mmio_pfn(pfn
)) {
1359 struct page
*page
= pfn_to_page(pfn
);
1360 if (!PageReserved(page
))
1364 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1366 void kvm_set_pfn_accessed(pfn_t pfn
)
1368 if (!kvm_is_mmio_pfn(pfn
))
1369 mark_page_accessed(pfn_to_page(pfn
));
1371 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1373 void kvm_get_pfn(pfn_t pfn
)
1375 if (!kvm_is_mmio_pfn(pfn
))
1376 get_page(pfn_to_page(pfn
));
1378 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1380 static int next_segment(unsigned long len
, int offset
)
1382 if (len
> PAGE_SIZE
- offset
)
1383 return PAGE_SIZE
- offset
;
1388 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1394 addr
= gfn_to_hva_read(kvm
, gfn
);
1395 if (kvm_is_error_hva(addr
))
1397 r
= kvm_read_hva(data
, (void __user
*)addr
+ offset
, len
);
1402 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1404 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1406 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1408 int offset
= offset_in_page(gpa
);
1411 while ((seg
= next_segment(len
, offset
)) != 0) {
1412 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1422 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1424 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1429 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1430 int offset
= offset_in_page(gpa
);
1432 addr
= gfn_to_hva_read(kvm
, gfn
);
1433 if (kvm_is_error_hva(addr
))
1435 pagefault_disable();
1436 r
= kvm_read_hva_atomic(data
, (void __user
*)addr
+ offset
, len
);
1442 EXPORT_SYMBOL(kvm_read_guest_atomic
);
1444 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
, const void *data
,
1445 int offset
, int len
)
1450 addr
= gfn_to_hva(kvm
, gfn
);
1451 if (kvm_is_error_hva(addr
))
1453 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1456 mark_page_dirty(kvm
, gfn
);
1459 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1461 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1464 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1466 int offset
= offset_in_page(gpa
);
1469 while ((seg
= next_segment(len
, offset
)) != 0) {
1470 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1481 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1484 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1485 int offset
= offset_in_page(gpa
);
1486 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1489 ghc
->generation
= slots
->generation
;
1490 ghc
->memslot
= gfn_to_memslot(kvm
, gfn
);
1491 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, gfn
, NULL
);
1492 if (!kvm_is_error_hva(ghc
->hva
))
1499 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1501 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1502 void *data
, unsigned long len
)
1504 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1507 if (slots
->generation
!= ghc
->generation
)
1508 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
);
1510 if (kvm_is_error_hva(ghc
->hva
))
1513 r
= __copy_to_user((void __user
*)ghc
->hva
, data
, len
);
1516 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, ghc
->gpa
>> PAGE_SHIFT
);
1520 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
1522 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1523 void *data
, unsigned long len
)
1525 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1528 if (slots
->generation
!= ghc
->generation
)
1529 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
);
1531 if (kvm_is_error_hva(ghc
->hva
))
1534 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
1540 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
1542 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
1544 return kvm_write_guest_page(kvm
, gfn
, (const void *) empty_zero_page
,
1547 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
1549 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
1551 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1553 int offset
= offset_in_page(gpa
);
1556 while ((seg
= next_segment(len
, offset
)) != 0) {
1557 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
1566 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
1568 void mark_page_dirty_in_slot(struct kvm
*kvm
, struct kvm_memory_slot
*memslot
,
1571 if (memslot
&& memslot
->dirty_bitmap
) {
1572 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
1574 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
1578 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
1580 struct kvm_memory_slot
*memslot
;
1582 memslot
= gfn_to_memslot(kvm
, gfn
);
1583 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
1587 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1589 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
1594 prepare_to_wait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1596 if (kvm_arch_vcpu_runnable(vcpu
)) {
1597 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
1600 if (kvm_cpu_has_pending_timer(vcpu
))
1602 if (signal_pending(current
))
1608 finish_wait(&vcpu
->wq
, &wait
);
1613 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1615 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
1618 int cpu
= vcpu
->cpu
;
1619 wait_queue_head_t
*wqp
;
1621 wqp
= kvm_arch_vcpu_wq(vcpu
);
1622 if (waitqueue_active(wqp
)) {
1623 wake_up_interruptible(wqp
);
1624 ++vcpu
->stat
.halt_wakeup
;
1628 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
1629 if (kvm_arch_vcpu_should_kick(vcpu
))
1630 smp_send_reschedule(cpu
);
1633 #endif /* !CONFIG_S390 */
1635 void kvm_resched(struct kvm_vcpu
*vcpu
)
1637 if (!need_resched())
1641 EXPORT_SYMBOL_GPL(kvm_resched
);
1643 bool kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
1646 struct task_struct
*task
= NULL
;
1649 pid
= rcu_dereference(target
->pid
);
1651 task
= get_pid_task(target
->pid
, PIDTYPE_PID
);
1655 if (task
->flags
& PF_VCPU
) {
1656 put_task_struct(task
);
1659 if (yield_to(task
, 1)) {
1660 put_task_struct(task
);
1663 put_task_struct(task
);
1666 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
1668 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1670 * Helper that checks whether a VCPU is eligible for directed yield.
1671 * Most eligible candidate to yield is decided by following heuristics:
1673 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1674 * (preempted lock holder), indicated by @in_spin_loop.
1675 * Set at the beiginning and cleared at the end of interception/PLE handler.
1677 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1678 * chance last time (mostly it has become eligible now since we have probably
1679 * yielded to lockholder in last iteration. This is done by toggling
1680 * @dy_eligible each time a VCPU checked for eligibility.)
1682 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1683 * to preempted lock-holder could result in wrong VCPU selection and CPU
1684 * burning. Giving priority for a potential lock-holder increases lock
1687 * Since algorithm is based on heuristics, accessing another VCPU data without
1688 * locking does not harm. It may result in trying to yield to same VCPU, fail
1689 * and continue with next VCPU and so on.
1691 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
1695 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
1696 (vcpu
->spin_loop
.in_spin_loop
&&
1697 vcpu
->spin_loop
.dy_eligible
);
1699 if (vcpu
->spin_loop
.in_spin_loop
)
1700 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
1705 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
)
1707 struct kvm
*kvm
= me
->kvm
;
1708 struct kvm_vcpu
*vcpu
;
1709 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
1714 kvm_vcpu_set_in_spin_loop(me
, true);
1716 * We boost the priority of a VCPU that is runnable but not
1717 * currently running, because it got preempted by something
1718 * else and called schedule in __vcpu_run. Hopefully that
1719 * VCPU is holding the lock that we need and will release it.
1720 * We approximate round-robin by starting at the last boosted VCPU.
1722 for (pass
= 0; pass
< 2 && !yielded
; pass
++) {
1723 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1724 if (!pass
&& i
<= last_boosted_vcpu
) {
1725 i
= last_boosted_vcpu
;
1727 } else if (pass
&& i
> last_boosted_vcpu
)
1731 if (waitqueue_active(&vcpu
->wq
))
1733 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
1735 if (kvm_vcpu_yield_to(vcpu
)) {
1736 kvm
->last_boosted_vcpu
= i
;
1742 kvm_vcpu_set_in_spin_loop(me
, false);
1744 /* Ensure vcpu is not eligible during next spinloop */
1745 kvm_vcpu_set_dy_eligible(me
, false);
1747 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
1749 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1751 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
1754 if (vmf
->pgoff
== 0)
1755 page
= virt_to_page(vcpu
->run
);
1757 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
1758 page
= virt_to_page(vcpu
->arch
.pio_data
);
1760 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1761 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
1762 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
1765 return kvm_arch_vcpu_fault(vcpu
, vmf
);
1771 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
1772 .fault
= kvm_vcpu_fault
,
1775 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1777 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
1781 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
1783 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1785 kvm_put_kvm(vcpu
->kvm
);
1789 static struct file_operations kvm_vcpu_fops
= {
1790 .release
= kvm_vcpu_release
,
1791 .unlocked_ioctl
= kvm_vcpu_ioctl
,
1792 #ifdef CONFIG_COMPAT
1793 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
1795 .mmap
= kvm_vcpu_mmap
,
1796 .llseek
= noop_llseek
,
1800 * Allocates an inode for the vcpu.
1802 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
1804 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
);
1808 * Creates some virtual cpus. Good luck creating more than one.
1810 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
1813 struct kvm_vcpu
*vcpu
, *v
;
1815 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
1817 return PTR_ERR(vcpu
);
1819 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
1821 r
= kvm_arch_vcpu_setup(vcpu
);
1825 mutex_lock(&kvm
->lock
);
1826 if (!kvm_vcpu_compatible(vcpu
)) {
1828 goto unlock_vcpu_destroy
;
1830 if (atomic_read(&kvm
->online_vcpus
) == KVM_MAX_VCPUS
) {
1832 goto unlock_vcpu_destroy
;
1835 kvm_for_each_vcpu(r
, v
, kvm
)
1836 if (v
->vcpu_id
== id
) {
1838 goto unlock_vcpu_destroy
;
1841 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
1843 /* Now it's all set up, let userspace reach it */
1845 r
= create_vcpu_fd(vcpu
);
1848 goto unlock_vcpu_destroy
;
1851 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
1853 atomic_inc(&kvm
->online_vcpus
);
1855 mutex_unlock(&kvm
->lock
);
1858 unlock_vcpu_destroy
:
1859 mutex_unlock(&kvm
->lock
);
1861 kvm_arch_vcpu_destroy(vcpu
);
1865 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
1868 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1869 vcpu
->sigset_active
= 1;
1870 vcpu
->sigset
= *sigset
;
1872 vcpu
->sigset_active
= 0;
1876 static long kvm_vcpu_ioctl(struct file
*filp
,
1877 unsigned int ioctl
, unsigned long arg
)
1879 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1880 void __user
*argp
= (void __user
*)arg
;
1882 struct kvm_fpu
*fpu
= NULL
;
1883 struct kvm_sregs
*kvm_sregs
= NULL
;
1885 if (vcpu
->kvm
->mm
!= current
->mm
)
1888 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1890 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1891 * so vcpu_load() would break it.
1893 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_INTERRUPT
)
1894 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
1898 r
= vcpu_load(vcpu
);
1906 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
1907 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
1909 case KVM_GET_REGS
: {
1910 struct kvm_regs
*kvm_regs
;
1913 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
1916 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
1920 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
1927 case KVM_SET_REGS
: {
1928 struct kvm_regs
*kvm_regs
;
1931 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
1932 if (IS_ERR(kvm_regs
)) {
1933 r
= PTR_ERR(kvm_regs
);
1936 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
1940 case KVM_GET_SREGS
: {
1941 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
1945 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
1949 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
1954 case KVM_SET_SREGS
: {
1955 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
1956 if (IS_ERR(kvm_sregs
)) {
1957 r
= PTR_ERR(kvm_sregs
);
1961 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
1964 case KVM_GET_MP_STATE
: {
1965 struct kvm_mp_state mp_state
;
1967 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
1971 if (copy_to_user(argp
, &mp_state
, sizeof mp_state
))
1976 case KVM_SET_MP_STATE
: {
1977 struct kvm_mp_state mp_state
;
1980 if (copy_from_user(&mp_state
, argp
, sizeof mp_state
))
1982 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
1985 case KVM_TRANSLATE
: {
1986 struct kvm_translation tr
;
1989 if (copy_from_user(&tr
, argp
, sizeof tr
))
1991 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
1995 if (copy_to_user(argp
, &tr
, sizeof tr
))
2000 case KVM_SET_GUEST_DEBUG
: {
2001 struct kvm_guest_debug dbg
;
2004 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
2006 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2009 case KVM_SET_SIGNAL_MASK
: {
2010 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2011 struct kvm_signal_mask kvm_sigmask
;
2012 sigset_t sigset
, *p
;
2017 if (copy_from_user(&kvm_sigmask
, argp
,
2018 sizeof kvm_sigmask
))
2021 if (kvm_sigmask
.len
!= sizeof sigset
)
2024 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2029 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2033 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2037 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2041 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2047 fpu
= memdup_user(argp
, sizeof(*fpu
));
2053 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2057 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2066 #ifdef CONFIG_COMPAT
2067 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2068 unsigned int ioctl
, unsigned long arg
)
2070 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2071 void __user
*argp
= compat_ptr(arg
);
2074 if (vcpu
->kvm
->mm
!= current
->mm
)
2078 case KVM_SET_SIGNAL_MASK
: {
2079 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2080 struct kvm_signal_mask kvm_sigmask
;
2081 compat_sigset_t csigset
;
2086 if (copy_from_user(&kvm_sigmask
, argp
,
2087 sizeof kvm_sigmask
))
2090 if (kvm_sigmask
.len
!= sizeof csigset
)
2093 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2096 sigset_from_compat(&sigset
, &csigset
);
2097 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2099 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2103 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2111 static long kvm_vm_ioctl(struct file
*filp
,
2112 unsigned int ioctl
, unsigned long arg
)
2114 struct kvm
*kvm
= filp
->private_data
;
2115 void __user
*argp
= (void __user
*)arg
;
2118 if (kvm
->mm
!= current
->mm
)
2121 case KVM_CREATE_VCPU
:
2122 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2124 case KVM_SET_USER_MEMORY_REGION
: {
2125 struct kvm_userspace_memory_region kvm_userspace_mem
;
2128 if (copy_from_user(&kvm_userspace_mem
, argp
,
2129 sizeof kvm_userspace_mem
))
2132 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
, 1);
2135 case KVM_GET_DIRTY_LOG
: {
2136 struct kvm_dirty_log log
;
2139 if (copy_from_user(&log
, argp
, sizeof log
))
2141 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2144 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2145 case KVM_REGISTER_COALESCED_MMIO
: {
2146 struct kvm_coalesced_mmio_zone zone
;
2148 if (copy_from_user(&zone
, argp
, sizeof zone
))
2150 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
2153 case KVM_UNREGISTER_COALESCED_MMIO
: {
2154 struct kvm_coalesced_mmio_zone zone
;
2156 if (copy_from_user(&zone
, argp
, sizeof zone
))
2158 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
2163 struct kvm_irqfd data
;
2166 if (copy_from_user(&data
, argp
, sizeof data
))
2168 r
= kvm_irqfd(kvm
, &data
);
2171 case KVM_IOEVENTFD
: {
2172 struct kvm_ioeventfd data
;
2175 if (copy_from_user(&data
, argp
, sizeof data
))
2177 r
= kvm_ioeventfd(kvm
, &data
);
2180 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2181 case KVM_SET_BOOT_CPU_ID
:
2183 mutex_lock(&kvm
->lock
);
2184 if (atomic_read(&kvm
->online_vcpus
) != 0)
2187 kvm
->bsp_vcpu_id
= arg
;
2188 mutex_unlock(&kvm
->lock
);
2191 #ifdef CONFIG_HAVE_KVM_MSI
2192 case KVM_SIGNAL_MSI
: {
2196 if (copy_from_user(&msi
, argp
, sizeof msi
))
2198 r
= kvm_send_userspace_msi(kvm
, &msi
);
2202 #ifdef __KVM_HAVE_IRQ_LINE
2203 case KVM_IRQ_LINE_STATUS
:
2204 case KVM_IRQ_LINE
: {
2205 struct kvm_irq_level irq_event
;
2208 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
2211 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
);
2216 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
2217 if (copy_to_user(argp
, &irq_event
, sizeof irq_event
))
2226 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
2228 r
= kvm_vm_ioctl_assigned_device(kvm
, ioctl
, arg
);
2234 #ifdef CONFIG_COMPAT
2235 struct compat_kvm_dirty_log
{
2239 compat_uptr_t dirty_bitmap
; /* one bit per page */
2244 static long kvm_vm_compat_ioctl(struct file
*filp
,
2245 unsigned int ioctl
, unsigned long arg
)
2247 struct kvm
*kvm
= filp
->private_data
;
2250 if (kvm
->mm
!= current
->mm
)
2253 case KVM_GET_DIRTY_LOG
: {
2254 struct compat_kvm_dirty_log compat_log
;
2255 struct kvm_dirty_log log
;
2258 if (copy_from_user(&compat_log
, (void __user
*)arg
,
2259 sizeof(compat_log
)))
2261 log
.slot
= compat_log
.slot
;
2262 log
.padding1
= compat_log
.padding1
;
2263 log
.padding2
= compat_log
.padding2
;
2264 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
2266 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2270 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
2278 static int kvm_vm_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2280 struct page
*page
[1];
2283 gfn_t gfn
= vmf
->pgoff
;
2284 struct kvm
*kvm
= vma
->vm_file
->private_data
;
2286 addr
= gfn_to_hva(kvm
, gfn
);
2287 if (kvm_is_error_hva(addr
))
2288 return VM_FAULT_SIGBUS
;
2290 npages
= get_user_pages(current
, current
->mm
, addr
, 1, 1, 0, page
,
2292 if (unlikely(npages
!= 1))
2293 return VM_FAULT_SIGBUS
;
2295 vmf
->page
= page
[0];
2299 static const struct vm_operations_struct kvm_vm_vm_ops
= {
2300 .fault
= kvm_vm_fault
,
2303 static int kvm_vm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2305 vma
->vm_ops
= &kvm_vm_vm_ops
;
2309 static struct file_operations kvm_vm_fops
= {
2310 .release
= kvm_vm_release
,
2311 .unlocked_ioctl
= kvm_vm_ioctl
,
2312 #ifdef CONFIG_COMPAT
2313 .compat_ioctl
= kvm_vm_compat_ioctl
,
2315 .mmap
= kvm_vm_mmap
,
2316 .llseek
= noop_llseek
,
2319 static int kvm_dev_ioctl_create_vm(unsigned long type
)
2324 kvm
= kvm_create_vm(type
);
2326 return PTR_ERR(kvm
);
2327 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2328 r
= kvm_coalesced_mmio_init(kvm
);
2334 r
= anon_inode_getfd("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
2341 static long kvm_dev_ioctl_check_extension_generic(long arg
)
2344 case KVM_CAP_USER_MEMORY
:
2345 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2346 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2347 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2348 case KVM_CAP_SET_BOOT_CPU_ID
:
2350 case KVM_CAP_INTERNAL_ERROR_DATA
:
2351 #ifdef CONFIG_HAVE_KVM_MSI
2352 case KVM_CAP_SIGNAL_MSI
:
2355 #ifdef KVM_CAP_IRQ_ROUTING
2356 case KVM_CAP_IRQ_ROUTING
:
2357 return KVM_MAX_IRQ_ROUTES
;
2362 return kvm_dev_ioctl_check_extension(arg
);
2365 static long kvm_dev_ioctl(struct file
*filp
,
2366 unsigned int ioctl
, unsigned long arg
)
2371 case KVM_GET_API_VERSION
:
2375 r
= KVM_API_VERSION
;
2378 r
= kvm_dev_ioctl_create_vm(arg
);
2380 case KVM_CHECK_EXTENSION
:
2381 r
= kvm_dev_ioctl_check_extension_generic(arg
);
2383 case KVM_GET_VCPU_MMAP_SIZE
:
2387 r
= PAGE_SIZE
; /* struct kvm_run */
2389 r
+= PAGE_SIZE
; /* pio data page */
2391 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2392 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
2395 case KVM_TRACE_ENABLE
:
2396 case KVM_TRACE_PAUSE
:
2397 case KVM_TRACE_DISABLE
:
2401 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
2407 static struct file_operations kvm_chardev_ops
= {
2408 .unlocked_ioctl
= kvm_dev_ioctl
,
2409 .compat_ioctl
= kvm_dev_ioctl
,
2410 .llseek
= noop_llseek
,
2413 static struct miscdevice kvm_dev
= {
2419 static void hardware_enable_nolock(void *junk
)
2421 int cpu
= raw_smp_processor_id();
2424 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2427 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
2429 r
= kvm_arch_hardware_enable(NULL
);
2432 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2433 atomic_inc(&hardware_enable_failed
);
2434 printk(KERN_INFO
"kvm: enabling virtualization on "
2435 "CPU%d failed\n", cpu
);
2439 static void hardware_enable(void *junk
)
2441 raw_spin_lock(&kvm_lock
);
2442 hardware_enable_nolock(junk
);
2443 raw_spin_unlock(&kvm_lock
);
2446 static void hardware_disable_nolock(void *junk
)
2448 int cpu
= raw_smp_processor_id();
2450 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2452 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2453 kvm_arch_hardware_disable(NULL
);
2456 static void hardware_disable(void *junk
)
2458 raw_spin_lock(&kvm_lock
);
2459 hardware_disable_nolock(junk
);
2460 raw_spin_unlock(&kvm_lock
);
2463 static void hardware_disable_all_nolock(void)
2465 BUG_ON(!kvm_usage_count
);
2468 if (!kvm_usage_count
)
2469 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2472 static void hardware_disable_all(void)
2474 raw_spin_lock(&kvm_lock
);
2475 hardware_disable_all_nolock();
2476 raw_spin_unlock(&kvm_lock
);
2479 static int hardware_enable_all(void)
2483 raw_spin_lock(&kvm_lock
);
2486 if (kvm_usage_count
== 1) {
2487 atomic_set(&hardware_enable_failed
, 0);
2488 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
2490 if (atomic_read(&hardware_enable_failed
)) {
2491 hardware_disable_all_nolock();
2496 raw_spin_unlock(&kvm_lock
);
2501 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2506 if (!kvm_usage_count
)
2509 val
&= ~CPU_TASKS_FROZEN
;
2512 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2514 hardware_disable(NULL
);
2517 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2519 hardware_enable(NULL
);
2526 asmlinkage
void kvm_spurious_fault(void)
2528 /* Fault while not rebooting. We want the trace. */
2531 EXPORT_SYMBOL_GPL(kvm_spurious_fault
);
2533 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2537 * Some (well, at least mine) BIOSes hang on reboot if
2540 * And Intel TXT required VMX off for all cpu when system shutdown.
2542 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2543 kvm_rebooting
= true;
2544 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2548 static struct notifier_block kvm_reboot_notifier
= {
2549 .notifier_call
= kvm_reboot
,
2553 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
2557 for (i
= 0; i
< bus
->dev_count
; i
++) {
2558 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
2560 kvm_iodevice_destructor(pos
);
2565 int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
2567 const struct kvm_io_range
*r1
= p1
;
2568 const struct kvm_io_range
*r2
= p2
;
2570 if (r1
->addr
< r2
->addr
)
2572 if (r1
->addr
+ r1
->len
> r2
->addr
+ r2
->len
)
2577 int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
2578 gpa_t addr
, int len
)
2580 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
2586 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
2587 kvm_io_bus_sort_cmp
, NULL
);
2592 int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
2593 gpa_t addr
, int len
)
2595 struct kvm_io_range
*range
, key
;
2598 key
= (struct kvm_io_range
) {
2603 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
2604 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
2608 off
= range
- bus
->range
;
2610 while (off
> 0 && kvm_io_bus_sort_cmp(&key
, &bus
->range
[off
-1]) == 0)
2616 /* kvm_io_bus_write - called under kvm->slots_lock */
2617 int kvm_io_bus_write(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2618 int len
, const void *val
)
2621 struct kvm_io_bus
*bus
;
2622 struct kvm_io_range range
;
2624 range
= (struct kvm_io_range
) {
2629 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2630 idx
= kvm_io_bus_get_first_dev(bus
, addr
, len
);
2634 while (idx
< bus
->dev_count
&&
2635 kvm_io_bus_sort_cmp(&range
, &bus
->range
[idx
]) == 0) {
2636 if (!kvm_iodevice_write(bus
->range
[idx
].dev
, addr
, len
, val
))
2644 /* kvm_io_bus_read - called under kvm->slots_lock */
2645 int kvm_io_bus_read(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2649 struct kvm_io_bus
*bus
;
2650 struct kvm_io_range range
;
2652 range
= (struct kvm_io_range
) {
2657 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2658 idx
= kvm_io_bus_get_first_dev(bus
, addr
, len
);
2662 while (idx
< bus
->dev_count
&&
2663 kvm_io_bus_sort_cmp(&range
, &bus
->range
[idx
]) == 0) {
2664 if (!kvm_iodevice_read(bus
->range
[idx
].dev
, addr
, len
, val
))
2672 /* Caller must hold slots_lock. */
2673 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2674 int len
, struct kvm_io_device
*dev
)
2676 struct kvm_io_bus
*new_bus
, *bus
;
2678 bus
= kvm
->buses
[bus_idx
];
2679 if (bus
->dev_count
> NR_IOBUS_DEVS
- 1)
2682 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
2683 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
2686 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
2687 sizeof(struct kvm_io_range
)));
2688 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
2689 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
2690 synchronize_srcu_expedited(&kvm
->srcu
);
2696 /* Caller must hold slots_lock. */
2697 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
2698 struct kvm_io_device
*dev
)
2701 struct kvm_io_bus
*new_bus
, *bus
;
2703 bus
= kvm
->buses
[bus_idx
];
2705 for (i
= 0; i
< bus
->dev_count
; i
++)
2706 if (bus
->range
[i
].dev
== dev
) {
2714 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
2715 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
2719 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
2720 new_bus
->dev_count
--;
2721 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
2722 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
2724 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
2725 synchronize_srcu_expedited(&kvm
->srcu
);
2730 static struct notifier_block kvm_cpu_notifier
= {
2731 .notifier_call
= kvm_cpu_hotplug
,
2734 static int vm_stat_get(void *_offset
, u64
*val
)
2736 unsigned offset
= (long)_offset
;
2740 raw_spin_lock(&kvm_lock
);
2741 list_for_each_entry(kvm
, &vm_list
, vm_list
)
2742 *val
+= *(u32
*)((void *)kvm
+ offset
);
2743 raw_spin_unlock(&kvm_lock
);
2747 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
2749 static int vcpu_stat_get(void *_offset
, u64
*val
)
2751 unsigned offset
= (long)_offset
;
2753 struct kvm_vcpu
*vcpu
;
2757 raw_spin_lock(&kvm_lock
);
2758 list_for_each_entry(kvm
, &vm_list
, vm_list
)
2759 kvm_for_each_vcpu(i
, vcpu
, kvm
)
2760 *val
+= *(u32
*)((void *)vcpu
+ offset
);
2762 raw_spin_unlock(&kvm_lock
);
2766 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
2768 static const struct file_operations
*stat_fops
[] = {
2769 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
2770 [KVM_STAT_VM
] = &vm_stat_fops
,
2773 static int kvm_init_debug(void)
2776 struct kvm_stats_debugfs_item
*p
;
2778 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
2779 if (kvm_debugfs_dir
== NULL
)
2782 for (p
= debugfs_entries
; p
->name
; ++p
) {
2783 p
->dentry
= debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
2784 (void *)(long)p
->offset
,
2785 stat_fops
[p
->kind
]);
2786 if (p
->dentry
== NULL
)
2793 debugfs_remove_recursive(kvm_debugfs_dir
);
2798 static void kvm_exit_debug(void)
2800 struct kvm_stats_debugfs_item
*p
;
2802 for (p
= debugfs_entries
; p
->name
; ++p
)
2803 debugfs_remove(p
->dentry
);
2804 debugfs_remove(kvm_debugfs_dir
);
2807 static int kvm_suspend(void)
2809 if (kvm_usage_count
)
2810 hardware_disable_nolock(NULL
);
2814 static void kvm_resume(void)
2816 if (kvm_usage_count
) {
2817 WARN_ON(raw_spin_is_locked(&kvm_lock
));
2818 hardware_enable_nolock(NULL
);
2822 static struct syscore_ops kvm_syscore_ops
= {
2823 .suspend
= kvm_suspend
,
2824 .resume
= kvm_resume
,
2828 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
2830 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
2833 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
2835 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
2837 kvm_arch_vcpu_load(vcpu
, cpu
);
2840 static void kvm_sched_out(struct preempt_notifier
*pn
,
2841 struct task_struct
*next
)
2843 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
2845 kvm_arch_vcpu_put(vcpu
);
2848 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
2849 struct module
*module
)
2854 r
= kvm_arch_init(opaque
);
2858 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
2863 r
= kvm_arch_hardware_setup();
2867 for_each_online_cpu(cpu
) {
2868 smp_call_function_single(cpu
,
2869 kvm_arch_check_processor_compat
,
2875 r
= register_cpu_notifier(&kvm_cpu_notifier
);
2878 register_reboot_notifier(&kvm_reboot_notifier
);
2880 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2882 vcpu_align
= __alignof__(struct kvm_vcpu
);
2883 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
2885 if (!kvm_vcpu_cache
) {
2890 r
= kvm_async_pf_init();
2894 kvm_chardev_ops
.owner
= module
;
2895 kvm_vm_fops
.owner
= module
;
2896 kvm_vcpu_fops
.owner
= module
;
2898 r
= misc_register(&kvm_dev
);
2900 printk(KERN_ERR
"kvm: misc device register failed\n");
2904 register_syscore_ops(&kvm_syscore_ops
);
2906 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
2907 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
2909 r
= kvm_init_debug();
2911 printk(KERN_ERR
"kvm: create debugfs files failed\n");
2918 unregister_syscore_ops(&kvm_syscore_ops
);
2920 kvm_async_pf_deinit();
2922 kmem_cache_destroy(kvm_vcpu_cache
);
2924 unregister_reboot_notifier(&kvm_reboot_notifier
);
2925 unregister_cpu_notifier(&kvm_cpu_notifier
);
2928 kvm_arch_hardware_unsetup();
2930 free_cpumask_var(cpus_hardware_enabled
);
2936 EXPORT_SYMBOL_GPL(kvm_init
);
2941 misc_deregister(&kvm_dev
);
2942 kmem_cache_destroy(kvm_vcpu_cache
);
2943 kvm_async_pf_deinit();
2944 unregister_syscore_ops(&kvm_syscore_ops
);
2945 unregister_reboot_notifier(&kvm_reboot_notifier
);
2946 unregister_cpu_notifier(&kvm_cpu_notifier
);
2947 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2948 kvm_arch_hardware_unsetup();
2950 free_cpumask_var(cpus_hardware_enabled
);
2952 EXPORT_SYMBOL_GPL(kvm_exit
);